Nunchux undercuts fal on Nano Banana images and Seedance image-to-video; the two tie on HappyHorse and Wan, and fal alone sells raw GPU hours
Asked:
“who's their top 2 competiotrs - Founding GTM - Careers - Nunchux AI | Nunchux AI”
Asked (follow-up):
“compare https://www.nunchux.ai/pricing with fal's pricing https://fal.ai/pricing and tell me the differences”
As of 2026-11-01, first-party Nunchux and fal pricing pages yield 43 distinct price rows across 10 models listed on both platforms (93 raw rows after removing URL and share-page repetitions), plus fal’s 6-GPU serverless compute table and Nunchux’s credit documentation. Units are US dollars per second of generated video, per image, per token bundle, or per GPU-hour.
Like-for-like unit prices on seven representative configurations
NunchuxfalLabel on the right: who is cheaper and by how much; “tie” means identical list prices. fal Seedance 480p rate is fal’s own approximation from token pricing.
Nunchux lists Nano Banana 2 at $0.0603/image vs fal’s $0.08 at default 1K — about 24.6% lower; Nano Banana Pro is $0.1206 vs $0.15 (−19.6%) and $0.216 vs fal’s 2× 4K rate of $0.30 (−28%). nunchux.ai · fal.ai
Seedance 2.5 cuts both ways: Nunchux image-to-video at 480p is $0.103/sec vs fal’s ~$0.2205/sec (−53.3%), but Nunchux text-to-video at $0.231/sec sits slightly above that same fal rate. nunchux.ai · fal.ai
fal publishes a raw GPU-hour menu that Nunchux does not: six GPUs from RTX Pro 6000 ($4.00 list / $1.99 as-low-as per hour) to B300 ($12.99 / $5.99). fal.ai
Veo 3.1 and Kling rows are not directly comparable: configurations differ (e.g. Nunchux Veo 3.1 i2v $0.36/sec vs fal $0.20/sec at 720p/1080p and $0.40 at 4K; fal’s Kling V3 rows are 4K-mode). Resolution, audio, task and token formulas materially change prices. nunchux.ai · fal.ai
Business-model differences
Nunchux — prepaid credits for model output
Prepaid credits, 1 credit = $1.00, deducted per successful request by model, performance tier and output size (docs.nunchux.ai).
Preset $10 / $20 / $50 packages or custom top-ups of $5–$5,000 whole dollars; $5 minimum, no stated free credits.
Speed/value tiers on Nunchux-optimized models (Radical Speed and Radical Value for FLUX, Qwen, HiDream O1), plus negotiated and trial discounts.
Dedicated enterprise endpoints at discounted monthly reserved rates via sales — but no public raw GPU price table on the reviewed materials (nunchux.ai).
fal — model APIs plus a public GPU-hour menu
Usage-based: “you only pay for the computing power you consume”, hosted model APIs priced per output (fal.ai).
Serverless & compute pricing for custom deployments on its GPU fleet, billed per hour with a list price and a discounted “as low as” rate per GPU.
Headline discount example: H100 as low as $2.49/h vs $4.50/h list.
fal GPU fleet, per hour
GPU
VRAM
List
As low as
All matched-model price rows
43 distinct rows after deduplicating repeated URLs and share pages from the 93 extracted rows. Conditions column preserves each page’s stated caveats; rows with differing task, resolution or token formulas are not directly comparable across platforms.
Platform
Model
Variant / task
Price
Unit
Conditions
Source
Method: first-party pricing pages of nunchux.ai, docs.nunchux.ai and fal.ai, captured 2026-11-01. 93 extracted matched-model price rows deduplicated to 43 distinct platform–model–variant–price rows; 12 fal GPU pricing rows; 56 Nunchux pricing-doc rows. Prices are public list rates in USD per second of generated video, per image, per token bundle, or per GPU-hour; discounts, regions, audio and token formulas can change effective prices. Evidence snippets and some non-comparable configurations were cut for space. No single URL backs more than half the comparison rows.